Artificial Intelligence in Upper Gastrointestinal Endoscopy: Current Evidence, Practice, and Future Directions

Artificial intelligence (AI) is advancing upper gastrointestinal endoscopy (UGIE), ranging from manual, variable workflows to assisted, quality-controlled examinations. Computer-aided detection (CADe) and diagnosis (CADx) systems can help in identifying anatomy, reducing blind spots, and shifting random biopsies toward targeted sampling. Across Barrett's neoplasia, esophageal squamous cell carcinoma, early gastric cancer, and <i>Helicobacter pylori</i> assessment, deep learning models may match or approach expert performance while accelerating image review. Architectures spanning convolutional neural networks to reinforcement learning tools demonstrate high sensitivity and specificity for lesion detection, invasion-depth estimation, and characterization. However, routine adoption requires rigorous, prospective, multicenter validation; mitigation of dataset bias and domain shift; attention to false positives, alarm fatigue, and workflow design; and training that prevents over-reliance on AI. With human-in-the-loop oversight, interpretable outputs, and cost-conscious deployment, AI can standardize inspections, improve diagnostic confidence, enhance training, and deliver better patient outcomes overall without displacing clinical judgment. In this narrative review, we aim to summarize these recent advancements, discuss the performance of AI across key upper gastrointestinal applications, and critically evaluate the practical challenges and future directions for its clinical implementation.